Spurious Relationships; Controlled Comparisons.

In a research design it is imperative that we be certain that the relationship we identify is for the variables we specified only. In other words, we need to be certain that there are no lurking variablesin our analysis that may have affected our findings. Our findings must be with all other things being equal, thus we must rely on controlled comparisons.

Controlled comparisons keep the levels of all other variables constant and only changes the values of the independent variable in order to measure the effect on the dependent variable. You shouldn't concern yourself with the math behind controlled comparisons, only know that this is the only way to measure the effect the independent variable has on the dependent variable without other variables getting in the way.

To understand why this is neccessary, consider a common problem researchers encounter when they try to study differences in Democrats and Republicans. Suppose we know there is a relationship that specifies that men become Republicans and women become Democrats. Thus, gender has a significant effect on partisanship. Now suppose that our tests revealed a strong positiverelationship between partisanship and opinions of welfare spending. Can we be sure that it is actually partisanship, and not another variable (like gender), that is causing welfare spending opinions? No, we can't be sure.

In this case we need to re-run the tests and control for gender. This will keep the amount of men and women in both groups (Republicans and Democrats) the same, thus measuring only the effect of partisanship on welfare spending opinions. Suppose we re-ran the tests and controlled for gender, and our results showed no relationship between partisanship and welfare spending opinions. What does that mean? It means that the original relationship (before controlling for gender) is spurious - it doesn't really exist. After controlling for gender the relationship disappeared. Gender is the variable with a strong relationship with welfare spending opinions, and since women are more likely to be Democrats the effect also showed up in our tests between partisanship and welfare spending opinions. Utilizing controlled comparisons is the only way to protect against having your tests effected by lurking variables. In module 5, you will learn much more about spurious relationships and other problems that lurking variables can cause.
-Randy Owen